GANsDTA: Predicting Drug-Target Binding Affinity Using GANs

Abstract

The computational prediction of interactions between drugs and targets is a standing challenge in drug discovery. State-of-the-art methods for drug-target interaction prediction are primarily based on supervised machine learning with known label information. However, in biomedicine, obtaining labeled training data is an expensive and a laborious process. This paper proposes a semi-supervised generative adversarial networks (GANs)-based method to predict binding affinity. Our method comprises two parts, two GANs for feature extraction and a regression network for prediction. The semisupervised mechanism allows our model to learn proteins drugs features of both labeled and unlabeled data. We evaluate the performance of our method using multiple public datasets. Experimental results demonstrate that our method achieves competitive performance while utilizing freely available unlabeled data. Our results suggest that utilizing such unlabeled data can considerably help improve performance in various biomedical relation extraction processes, for example, Drug-Target interaction and protein-protein interaction, particularly when only limited labeled data are available in such tasks. To our best knowledge, this is the first semi-supervised GANs-based method to predict binding affinity.

DeepH-DTA: Deep Learning for Predicting Drug-Target Interactions: A Case Study of COVID-19 Drug Repurposing

“…On the other hand, deep-learning techniques that automatically capture feature representation show great performance improvement. The * denote the proposed architecture The * denote the proposed architecture First, this paper considers a few recent textual representation approaches such as: DeepDTA [29], MT-DTI [32], Deep-CPI [28], WideDTA [7], GANsDTA [13], and Attention-DTA [14]. Among these approaches, Attention-DTA and MT-DTI yielded best results with CI of 0.887, MSE of 0.245 on the Davis dataset; also, on the KIBA dataset, they both achieved CI of 0.882 and MSE of 0.220 and 0.162 respectively.…”

“…In [7], Ozkirimli et al introduce a methodology for predicting DT binding affinities using CNN over word representation of protein and compound sequences demonstrating that most essential binding information implanted in the protein domain. Zhao et al [13] proposed a generative adversarial network (GAN) to learn beneficial patterns within labeled and unlabeled sequences and utilized convolutional regression to forecast binding affinity score. However, they did not address GAN training on a small dataset.…”

“…To tackle this problem, we expand node-level attention to multi-head attention to preserve the stability of the training process. We apply the node-level attention to times, and the learned embeddings are concatenated to obtain the semantic-specific embedding as formulated in equation (13),…”

“…Thus, we select values of 7 and 12.1 as a threshold for the Davis and KIBA datasets, respectively. We choose these values based on because of its proven optimality and wide adoption in previous studies [7,29,13].…”

Autoencoders for Drug-Target Interaction Prediction

“…Zhao et al [18] proposed an end-to-end model, associated with an attention mechanism, to predict the binding affinity of DTIs. Zhao et al [19] proposed a neural network, GANsDTA, which combined two Generative Adversarial Networks (GANs) and a regression network to predict binding affinity. Those deep learning models show great promise for identifying DTIs.…”